Immigration Is Not One Thing That Has Effects

Deliberate policy changes, not varied abstractions, drive outcomes we care about

A lot of political debate about immigration is a fight over its “effects.” On one side, opponents warn that immigration depresses wages, raises crime, and erodes social cohesion. On the other side, advocates insist that immigration boosts growth, does not increase crime, and might even lower it. Both sides, especially the generally more educated pro-immigration side, like to say that “the evidence” is on their side, and they can point to studies, figures, and even regression models to back it up. What I would like to argue here is that both sides… are wrong.

Hear me out. I’m increasingly convinced that, regardless of your empirical convictions, the claims about positive or negative effects of “immigration” (and more abstract concepts like “inequality” or “diversity”) are vastly misleading. And I’m not just trying to be a contrarian here or knee-jerk centrist. When someone says that we should acknowledge that immigration has a mix of good and bad effects, this is slightly better…but also wrong.

Of course, there is a large, careful academic literature on the relationship between immigration and things like wages and crime, usually pointing to neutral or positive effects. Lauren Gilbert has done a great series of live posts summarizing a huge chunk of it, which I highly recommend checking out. For quite some time, though, something has been bothering me about the idea of figuring out the “effects” of big concepts like immigration in the first place. The recent post by Tibor Rutar on whether inequality undermines democracy, and the discussion of the likely inflated role of surveys in the credibility revolution helped me pin this down a bit more.

So, what I’ll try to convince you here is that “immigration” in the abstract cannot and does not have any true, or at least practically true, identifiable effects on relevant public outcomes. This is because “immigration” is not a decision or a policy lever that can be manipulated deliberately. It is not even one single specific thing. It is a description of different people moving between different places under different rules.

My point is not just about semantics. It is about what we can realistically know and what we can practically change. The problem with talking as if “immigration” has some true effects we need to uncover is that it hides the actual decisions that states make about who can come, on what terms, and what happens after they arrive. Those are the things that have effects, because people and governments can intentionally change them. Some immigration policies and decisions can and do produce better or worse outcomes than others. The question is which ones.

Selectivia and Inclusivia: a tale of two states

To see why this matters, it helps to start with a simple thought experiment. Indulge me for a moment while I tell you a story of two (totally made-up) countries. Call them Selectivia and Inclusivia.

Both are wealthy democracies with similar economies where people speak the same language. Both receive roughly the same number of culturally different immigrants per capita every year from the exact same poorer region. If you just looked at their net migration rates and the immigrant origin, you wouldn’t expect to see much difference in relevant outcomes.

But, for various reasons, their governments decided to run their immigration systems very differently. Selectivia uses a very demanding points-based system. It mostly admits highly educated workers with strong language skills and job offers in productive sectors. It vets people carefully and enforces the rules. A serious crime can get you deported. Employers who hire unauthorized workers face real penalties.

Inclusivia is also rich, but does not want to be seen as “picky” because it preaches inclusivity (duh!). It relies heavily on humanitarian admissions and family reunification, while keeping its work immigration pathways tight. It allows long backlogs for work visas, prohibits asylum seekers from working legally (for their own good), but also enforces these rules weakly. Many newcomers end up in informal jobs and in ethnic neighborhoods with little external support.

On paper, both countries have “high, ethnically diverse immigration.” In practice, they are moving different people into different legal and economic environments. Now imagine the best scholars of Selectivia and Inclusivia try to answer the question like “Does immigration increase crime?” in their countries using the best administrative data and quasi-experimental identification strategies.

In Selectivia, the immigrants are heavily screened, quickly employed, and know that any serious offense can get them deported. You would not be surprised to find that they commit fewer crimes than comparable natives. In Inclusivia, they bring in a lot of young men with few legal job options, so you might find higher crime rates among immigrants. Even if you only care about what’s happening in only one of these countries, however, it would clearly be a mistake to conclude that “immigration” is inherently good or bad for public safety.

Will everything be different across the two countries? No. In both places, for instance, scholars will likely find that immigrants are more likely to speak different languages than natives. That is just part of what it means to be an immigrant. But for outcomes we care about politically — wages, fiscal contributions, crime — what matters is not “immigration” in the abstract. It is how the system is designed and who it brings in.

Of course, even policy effects depend on the context. The UK, for instance, has tried to adapt Australia’s points system several times but has made a mess of it. Copy-pasting a law does not copy-paste its effects. Yet there is still a meaningful difference between saying “immigration increases crime” and saying “banning refugees from work for a year after arrival tends to increase crime.” The former is mostly an abstract political slogan. The latter is something we can identify, argue about, and feasibly change.

What do people mean when they say “immigration affects crime”?

The International Organization for Migration defines immigration as “the act of moving into a country other than one’s country of nationality or usual residence.”¹ That is a description of a spatial demographic process determined by many push and pull factors, not a single quantity. There is no world government or any other entity that could turn “the immigration dial” up or down.

When ordinary people say “immigration raises crime,” they often have in mind either more incidents of crime in absolute terms or a very particular image of certain foreigners committing crimes as broadcast by the media. When advocates say “immigration does not raise crime,” they often mean that immigrants commit crimes at lower rates than native-born citizens, which is true in some but not all contexts.

When researchers talk about whether one thing “causes” another, they usually have in mind a counterfactual statement or something like Judea Pearl’s “do-operator”: what would happen if we changed X while holding everything else about the world fixed. The cleanest way to do this is a randomized controlled trial: you give the treatment to the experimental group and nothing (or a placebo) to the control group, and then compare the outcomes. Unfortunately, figuring out what would happen if we increased or decreased “immigration” is more complex than that.

Social scientists often use a variety of designs: looking at changes after a certain shock like in the case of the famous Mariel Boatlift study or “shift-share instrumental variables” by looking at pre-existing ethnic networks. In the real world, however, the only way to apply the “do-operator” to immigration is through deliberate policy and decision-making. Democratic governments can decide to change visa caps, eligibility criteria, enforcement practices, or rights after arrival, and those changes lead to different numbers and types of people moving.²

Some economists like James Heckman call the impact of a specific change in the rules or decisions a “policy-relevant treatment effect.” Given this framework, instead of asking “What is the effect of immigration on crime?” we should rather ask “What is the effect of admitting this particular group of workers under this visa program in this period on crime?” That sounds more narrow, but it is the only kind of “immigration-related effect” we can actually identify.

The problem is not just a lack of nuance

Once you see this, a lot of the public debate looks strange. In the United States, the most rigorous studies tend to find that all possible types of immigrants, on average, commit less crime than natives and that plausibly exogenous changes in immigration are not associated with higher crime rates. European evidence is much more mixed, in part because the migrant populations and enforcement environments are very different.³

Yet pro-immigration scholars and advocates often jump from these findings to more sweeping claims about the effects of immigration. As a result, you have people in Europe confidently declaring that science proves “immigration does not raise crime” by citing US studies and data as if this is some kind of ground truth. At the same time, as Matt Burgess has recently pointed out, you also often have American restrictionists citing European crime data, as if a spike in offenses among poorly integrated refugee youth in Sweden tells you what will happen if you expand high-skill work visas to more Indians in the US.

Migration is what you make it. If your policy selects educated professionals, screens them for criminal histories, and makes any serious offense grounds for deportation, you should expect very low crime rates among immigrants. If your policy strands young men with no legal way to work, in marginalized neighborhoods, with little support and weak enforcement, you should expect more crime. Both of these systems can be called “immigration.” Neither tells you what “immigration” as such does.

I am sympathetic to the argument that we cannot always acknowledge nuance in public communication. If you attach every caveat you know to every sentence, nobody will finish your op-ed. But the problem here is not too much nuance. It is the wrong level of abstraction.

When we say “immigration lowers crime,” we are not bravely cutting through complexity. We are skipping over the only levers we actually control — policies — and pretending there is a single, context-free object called “immigration” whose true effect we just have to discover.

Policies are the levers that move outcomes

So what does have effects? Here I think the “credibility revolution” in economics and political science has the right basic idea. The most useful studies are not the ones that try to estimate some grand, policy-free effect of immigration in general. They are the ones that cleanly identify the consequences of a specific, realistic change in policy for the people on the margin of that policy. Think of governments raising or lowering a visa cap, changing a work authorization rule, or tightening enforcement in some domain.

Even if we had perfect data and unlimited resources, we could not say “the true effect of immigration on crime is X.” There is no single parameter to estimate. Any credible claim will always be about a particular policy change for a particular group in a particular environment. But it should be possible to say that “the effect of introducing language programs for immigrants across contexts is on average Y.”

Michael Clemens and Ethan Lewis offer a nice example of such “policy-relevant treatment effects” in their study of low-skill work visas in the US. Instead of asking like many whether “immigration helps or hurts US workers,” they use the H-2B visa lottery — a randomized cap on low-skill non-farm guest workers — to compare similar firms that randomly win and lose access to those workers. They find that firms allowed to hire more H-2B workers expand production and investment, with no evidence of overall job losses for US workers and possible gains in some areas.

You can say that this is a fact about “immigration.” But, more specifically, it is a fact about what happens when you let US firms hire more low-skill seasonal workers, legally, through a particular program, in the 2021 H-2B visa lottery and its aftermath. That is exactly the kind of thing policymakers actually decide and care about. Of course, when we have many studies like that with converging findings, we can generalize more. But it will ultimately be generalization about policies regulating immigration, not necessarily immigration in itself.

Interlude: inequality and diversity are not things that have effects either

If you still are not convinced, I hope you can at least grant the possibility that the same problem must arise with even more abstract concepts like “inequality” and “diversity.” When I first took a social science class as an undergraduate, I was struck by how much of sociology and adjacent disciplines are organized around inequality and disparities between various groups. Not just as possible outcomes of interest but causes of other important things like democracy and violent conflict.

I still think both topics matter as important diagnostics or indicators to measure. But they are not levers in themselves either. Inequality is a summary of the income distribution. Diversity is a summary of the demographic mix. Neither jumps out of the data and changes your life on its own. The reason why some places are more (un)equal or diverse than others is never random and overdetermined by various interconnected factors. So, when smart people say “inequality hurts democracy” or “diversity undermines social trust” by pointing out to some cross-national correlations or even quasi-experimental studies, I’m really not quite sure how to interpret those statements.

As I show in my research with Giuliana Pardelli using new data from Brazil, for instance, many of the observed negative (or positive) effects of local ethnic diversity are a result of a statistical artifact related to historical state development and the incentives of certain populations to settle in more remote areas. There are some historical examples of authoritarian governments purposefully resettling entire ethnic groups, but I hope it is not something anyone would want to follow or replicate.

Ethnic diversity as an outcome: the coevolution of state capacity and racial demography in Brazil

At the same time, if governments decide to reduce measured inequality by confiscating the wealth of everyone in the top 1 percent and throwing it into the sea (or some ineffective program), you will get very different “effects” than if you reduce inequality by raising the incomes of the bottom half via, let’s say, cash transfers. Both will show up as a lower Gini coefficient. Only one will be a (relative) success for human well-being.

So, what is to be done?

While “migration” understood as a description of people moving between places does not have a single and true identifiable effect, the way we regulate it likely does. It is fine to use shorthand like “immigration raises productivity” or “immigration does not increase crime” in a headline. But as researchers and commentators, we should not confuse our own shortcuts with ground truth.

Importantly, the point is not that we should prioritize niche, well-identified RCTs over larger descriptive studies or even theoretical modeling. Good, relevant evidence is not just about clever causal identification. It is about tying that identification to a concrete policy lever. Instead of asking “Is immigration good or bad?”, we should ask things like “If we change these visa allocations in this way, what happens to innovation, wages, and public opinion?”

That has two practical implications. First, we should be much more explicit about what we are actually talking about, at least when it comes to technical reports and academic papers. Instead of saying “immigration reduces crime,” say “in recent US data, legal immigrants with strong labor market ties and deportation risk commit fewer crimes than comparable natives.” It is more cumbersome, but it is honest and actually useful.

Second, in both advocacy and analysis, we should force ourselves to think in terms of better and worse policies, not better and worse “amounts of immigration.” If you are an advocate, ask yourself not just how immigration can be good, but how it can realistically be bad, and under what rules. If you are a skeptic, ask yourself what concrete system you would support to be convinced that certain migrants improve public services, pay more in taxes than they receive in benefits, and reduce crime in certain areas. What admissions criteria, enforcement practices, and integration policies would make those outcomes more likely?

As I recently told Kelsey Piper for her Argument piece, I have no patience for people who claim that we do not need to change immigration policy because “all serious studies show immigration is already beneficial.” Or that acknowledging that some immigration policies have bad outcomes plays into right-wing xenophobic narratives.

The policy status quo in most OECD countries is far from optimal. In a market economy, prohibiting or heavily discouraging people from moving to where they are most productive is bad for them and bad for everyone else. It is hard to justify morally. Yet that still does not mean that “immigration” in the abstract is good under any regime, or that we can ignore the design of the system.

Migration is basically just a fancy word to describe people moving, which they have been doing for centuries. Bad policies turn those movements into various losses, but good policies can turn them into gains. The effects of better regulations, not “immigration” as some single metaphysical object of nativist fear or humanitarian sacred right, is what we should be arguing about more.

1. Some definitions also specify that the term “immigration” is only reserved for “permanent settlement” but this distinction is rather irrelevant for the purposes of this post.

2. Nonprofits can also decide to fund an integration program in line with government rules, and individuals can decide to support certain political parties or nonprofits based on their immigration preferences.

3. Lauren Gilbert’s excellent reviews on immigration and crime in the United States and immigration and crime in Europe make this case clearly.